478 



Fishery Bulletin 100(3) 



observed variables (species abundances) were correlated. 

 Each PC represents a particular trend in habitat use, and 

 the axis differentiates among sets of species that are likely 

 to be found in contrasting conditions for variables that 

 define the PC. Because the PCs define axes, it is possible to 

 plot species on these axes (using the PC scores) and thus 

 identify those with contrasting or similar patterns of habi- 

 tat use. Species with intermediate scores might be char- 

 acteristic of intermediate environments, or they might be 

 found over the entire range of environments reflected by a 

 PC (because a species' score is the weighted mean of scores 

 of samples where it was found). Habitat associations by 

 demersal nekton were investigated separately from sites 

 in the north and south because previous observations 

 have suggested that areas to the north and south of the 

 Orange River are distinct (Field et al., 1996). 



Habitat selection by individual species can be deter- 

 mined by comparing species' variances on each PC with 

 variance of the environment. Environmental variance on 

 a PC was determined by calculating PC scores for each 

 sample, following standardization with the appropriate 

 mean of means, and standard deviation of means (see Fel- 

 ley and Vecchione, 1995). Then the scoring function was 

 applied to each sample and the variance determined as 

 the variance of sample scores. The score of each sample 

 was assigned to each individual of all species seen in that 

 sample. A species' variance was then calculated for each 

 species as the variance of these scores. Levene's test was 

 used to compare the species' variance with environmental 

 variance. The null hypothesis invoked for these tests was 

 "species" variance is not significantly different from en- 

 vironmental variance with respect to each factor." Active 

 habitat selection by a species was inferred when a species 

 variance was significantly smaller than the observed en- 

 vironmental variance ( 1-tailed test). This implies that the 

 species was actively selecting a subset of the available en- 

 vironment with respect to that PC, i.e. species distribution 

 in the study area was not random (see Felley et al., 1989). 

 Comparisons of species variances with sample variances 

 on each axis were performed only on the most common 

 species, in each area because the sample sizes for most 

 species were too small to allow statistically meaningful 

 tests. 



Assemblages Principal component analysis of the data 

 revealed how species were distributed along composite 

 environmental axes. Although species that are distributed 

 in a similar way on a PC axis can be interpreted to share 

 common responses to that axis, they do not necessarily 

 occur together in the same assemblage. Alternative meth- 

 ods of analysis for examining species associations directly 

 can be used to determine assemblage composition; as 

 an alternative method, the software package PRIMER 

 (Clarke and Warwick, 1994) was used in our study. With 

 this software, descriptive, multivariate statistics allowed 

 an examination of the relationships among the samples 

 (based on similarities in specific composition of nekton) 

 to determine how they were distributed with respect 

 to the physical environment. Abundance of only those 

 species that occurred in greater than 5% of the samples 



were root-root transformed, and a similarity matrix was 

 constructed between the samples containing two or more 

 species by using the Bray-Curtis index (Field et al., 1982). 

 These matrices were used to plot classification diagrams 

 of percentage similarity between samples by means of 

 group-average sorting (fuller details of this method are 

 provided by Field et al., 1996). Owing to the small area 

 filmed in individual videotape samples (mean 26.20 m^, 

 SD 11.24 m-), and the generally low density of most fish 

 species, the samples were pooled (as in Dennis and Bright, 

 1988) by the nature of the substratum within a dive (hard, 

 soft, mixed), so that replicates were of dives and not of 

 individual videotape samples. 



Activities In an effort to understand the behavior of 

 demersal fishes in their natural environment, the activity 

 of each demersal fish (on first sighting) was assigned to 

 one of the following behavior patterns: 



1 hovering off the substratum; 



2 positioned on the substratum; 



3 swimming in the water column; 



4 positioned in a crevice or under an overhang; 



5 occupying a shelter hole; 



6 buried (fully or partially) in the substratum. 



Observations of the substratum over which the fish 

 was seen were also recorded, in the hope that it would be 

 possible to correlate behaviors with the environment. The 

 activity of each of the dominant species of demersal fishes 

 was analyzed by using percent occurrence. This value was 

 calculated by dividing the sum of all individuals observed 

 in each activity (per major substratum type) by the total 

 number of individuals of the species for which activities 

 were recorded. 



Results 



Associations with environmental features 



A total of 22 different taxa of demersal nekton were identi- 

 fied from the samples (Table 2). Eight of these were seen 

 only once or twice, and only hakes occurred in more than 

 50'S of the samples. The hakes were assumed to be Mer- 

 luccius capensis on the basis of their inshore distribution 

 (Roel, 1987). It was not always possible to clearly separate 

 gobies, Sufflogobius bibarbatus, from dragonets, Paracal- 

 lionymus costatus, and all cases of ambiguity were elimi- 

 nated (i.e. only those fish that could be identified to species 

 were included in the analyses). 



Each species tended to be associated with slightly differ- 

 ent features of the physical or biotic environment (or both). 



North Eleven species were seen among these samples, 

 and their mean environments are given in Table 3. PCA 

 of the correlation matrix, generated from the mean envi- 

 ronments of the six dominant species of nekton, produced 

 three PCs with eigenvalues greater than one (Table 4). 

 PCI alone explained more than 60% of the variance in the 



